CISA added CVE-2026-97229 to the Known Exploited Vulnerabilities catalog on 29 October, with a federal remediation date of 1 November and a required action that sends agencies to BOD 26-04 and the Forensics Triage Requirements 1. GitLab shipped the fix seventeen days earlier, on 12 October, in 19.4.2, 19.3.5 and 19.2.8 2. Neither fact tells you whether a particular server is safe.
The part I keep rereading sits in a short incident writeup that Alpine Linux published on 28 October 3. They ran 19.3.4 through an attack on 20 and 21 October, upgraded to 19.3.6 on 22 October, and their own check on 26 October found the instance still vulnerable, because the patch they needed sat in the Ruby build under GitLab and their image carried standard Ruby. They closed the service, rebuilt the instance from scratch, reset every password and replaced the secret keys.
What a 10.0 requires
GitLab scored the flaw 10.0 with a vector that reads no privileges required and no user interaction 2. Rapid7 reproduced code execution on an unpatched 19.4.1 CE image, published the proof of concept, and described the condition the score buries: the attacker needs to plant a .gitlab-ci.yml and a repository path with UTF-8 characters in a public project 4. An ordinary account gets there on any instance that lets users create public projects, and Rapid7 notes that self-registration removes that constraint 4.
That changes my first question. I want to know who can create a public project on that server this morning, before anyone quotes me a version number. An instance with self-registration off and public project creation restricted has a narrower path than the score implies, and an instance that hands out accounts to anyone has the whole path.
Rapid7 lists the interim controls too, and names the limit on them. Restricting anonymous access to public projects and stopping untrusted users from creating or writing to public projects can reduce exposure to the demonstrated delivery path. Neither control remediates the compiler 4.
The overflow lands after validation
The mechanism moves the fix off the product version. A crafted pattern compiles in a compact US-ASCII form and passes GitLab’s regular expression validation. On the second pass, against a valid UTF-8 repository path, Ruby’s Onigmo engine recompiles it for the new encoding after those checks have finished. An integer overflow in the program-size calculation produces invalid instructions, and the resulting loss of control flow lets an attacker run their own Onigmo operations inside the Ruby process and execute an operating system command as the GitLab git user 4.
Ruby shipped the engine fix on 23 October in 3.3.13, 3.4.12 and 4.0.9 5. That advisory also rules out two mitigations people reach for first. Regexp.timeout does not protect against these bugs, and Ruby calls length limits unreliable protection, because some patterns trigger with short input 5. The same page lists four GitLab CVEs from this engine work, two fixed in 19.4.1 and one more in 19.5.1 5. I read that as a series, which is why the Ruby version is now part of how I track this server.
What the Alpine week shows
Their timeline is what I would put in front of a change board, because each step is a decision someone could make again next month 3. A newly registered account sent the crafted CI configurations. GitLab worker processes crashed eight times, and in four of those crashes the attacker got partway through and failed at the next step. The attacker downloaded their GitLab container image, which reads like tuning an exploit against a specific build. The 22 October upgrade to 19.3.6 did not clear the vulnerability, and they kept the instance reachable until 27 October, when they shut it to everyone except developers and build servers.
In the writeup they say a successful exploit would leave no crash and no trace in the web logs 3. They could not rule out code execution, partly because some logs from the attack day no longer existed and partly because they recorded no outbound connections from that host. A crash-free success is the case my checks are least ready for.
The four things I hand over
The server is not mine to patch, so my part is to answer the version question and hand the ordering and the evidence to whoever owns it.
First, which Ruby. If the instance runs GitLab’s own package or image, one of the fixed releases (19.2.8, 19.3.5, 19.4.2) carries the patched Ruby. If it runs a distribution image or a build with its own Ruby, the GitLab version number says nothing about this flaw, and Ruby has to be 3.3.13, 3.4.12, 4.0.9 or later 3. That check runs in one line on the host, and it tells me more than the release page does.
Second, the exposure path. Three settings decide whether the demonstrated chain stays open on that host: self-registration, public project creation, and anonymous access to public projects 4.
Third, detection. GitLab published six detections alongside the release, and their names describe signals a monitoring team can implement: an Onigmo native worker crash, a CI lint workhorse worker crash, an oversized non-ASCII path in a CI regexp, a CI regexp guard rejection, a burst of scripted client errors against CI lint, and a hostile regexp against the GraphQL CI lint path 2. Crash signals miss the clean success, so I pair them with a look for unexpected processes and outbound connections on the host.
Fourth, triage. The KEV required action asks for forensics triage, which turns a version report into a statement of what we looked for and what we found 1. The Alpine writeup shows how little of that evidence survives a week of incident work, with the old container’s filesystem and some background job logs gone.
The order
Ruby proof first, because it decides whether a patch exists on that host at all. Exposure second, since it sets how loud the message needs to be. Detection third: the window has been open since 12 October whether or not anyone used it. Triage last, because it is the only item that answers a question someone will ask after the fact.
Their instance ran 19.3.6 while the Ruby underneath stayed unpatched, and by the time they caught it the only remedy they could stand behind was a full rebuild. I would rather spend an hour proving which Ruby a server carries than a week working out what ran on it.
-
National Institute of Standards and Technology. (2026, October 30). CVE-2026-97229 detail. National Vulnerability Database. https://nvd.nist.gov/vuln/detail/CVE-2026-97229 ↩︎ ↩︎
-
GitLab Inc. (2026, October 12). GitLab critical patch release: 19.4.2, 19.3.5, 19.2.8. GitLab Docs. https://docs.gitlab.com/releases/patches/patch-release-gitlab-19-4-2-released/ ↩︎ ↩︎ ↩︎
-
Alpine Linux. (2026, October 28). GitLab security incident: attempted exploitation of CVE-2026-97229. Alpine Linux. https://alpinelinux.org/posts/GitLab-security-incident-CVE-2026-97229.html ↩︎ ↩︎ ↩︎ ↩︎
-
Rapid7. (2026, October 13). CVE-2026-97229: Critical unauthenticated remote code execution in GitLab. Rapid7 Blog. https://www.rapid7.com/blog/post/etr-cve-2026-97229-critical-unauthenticated-remote-code-execution-in-gitlab/ ↩︎ ↩︎ ↩︎ ↩︎ ↩︎
-
k0kubun. (2026, October 23). Regexp compilation fixes related to GitLab’s CVE-2026-89078, CVE-2026-93577, CVE-2026-97229 and CVE-2026-110264. Ruby. https://www.ruby-lang.org/en/news/2026/10/23/regexp-fixes-for-gitlab-cves/ ↩︎ ↩︎ ↩︎